Centro de Estudios Distributivos, Laborales y Sociales · Distributivos, Laborales y Sociales Maestría en Economía Facultad de Ciencias Económicas Conditional Cash Transfers, Payment

Este documento de trabajo es parte del proyecto Mercados de Trabajo para el Crecimiento Inclusivo en América Latina del Centro de Estudios Distributivos, Laborales y Sociales (CEDLAS) con el apoyo del Centro Internacional de Investigaciones para el Desarrollo (IDRC-Canadá).

CC | EE | DD | LL | AA | SS

Centro de Estudios Distributivos, Laborales y Sociales

Maestría en Economía

Facultad de Ciencias Económicas

Conditional Cash Transfers, Payment Dates and Labor Supply: Evidence from Peru

Fernando Fernandez y Victor Saldarriaga

Documento de Trabajo Nro. 140

Enero, 2013

ISSN 1853-0168

Conditional Cash Transfers, Payment Dates and Labor Supply:Evidence from Peru∗

Fernando Fernandez†

Victor Saldarriaga‡

November 19, 2012

Abstract

We assess the effects of a Conditional Cash Transfer program on adult labor supplyin Peru. The program, named Juntos, lacks an experimental design so we rely on a sort of“natural experiment”. Instead of comparing treated and non-treated households, our strategyexploits within-municipality variation in the distance between payment dates of Juntos andinterview dates of the Peruvian National Household Survey. We find that having receivedthe cash transfer two weeks before the interview causes a reduction of 6 hours of work ofrecipients during the week prior to the survey. These effects are larger for married womenand for mothers with children aged 5 or less. In addition, results are robust to differentspecifications and changes in the sample.

Keywords: Conditional Cash Transfers, Labor supply, Juntos, PeruJEL Codes: I38, J22

∗We are indebted to Leonardo Gasparini for guidance and support. Also, we thank Cesar Calvo, Viviana Cruzado,Rosamaria Dasso, Nicolas Dominguez, Hugo Ñopo, Ingo Outes, Renato Ravina, Alan Sanchez, Jose Valderrama and seminarparticipants at Universidad de Piura, the Peruvian Ministry of Economy and Finance, and the Central Bank of Peru forhelpful comments. We gratefully acknowledge financial support from the II Contest of Essays on Labor and Social Issuesin Latin America, organized by CEDLAS (UNLP-Argentina) and the IDRC (Canada). All remaining errors are ours.†Barcelona GSE and Universidad de Piura. [email protected]‡Peruvian Ministry of Social Development and Inclusion. [email protected]

1

1 Introduction

Around the world, Conditional Cash Transfers (henceforth, CCTs) are considered powerfulmeans to reduce current and future poverty. After the success of programs such as Progresa inMexico and Bolsa Escola in Brazil, almost every country in Latin America has implemented itsown program. There are many rigorous impact evaluations of CCTs but most of these studieshave paid little attention to the effects of cash transfers on the labor supply of adults in treatedhouseholds.

This paper analyzes the impact of a CCT program in Peru, known as Juntos, on adult laborsupply. Juntos lacks an experimental design so we require a credible identification strategy.In particular, we exploit differences in the interview dates of the National Household Surveyand the payment dates of the program within a given municipality. The timing of the interviewcombined with the payment schedule of Juntos generates a sort of “natural experiment” in whichsome households are interviewed just after the payment while others are surveyed weeks lateror before. We find that if the cash transfer occurs two weeks before the interview, recipients’hours of work are reduced by 6 hours during the week prior to the survey. This reduction israther large since it implies a fall of roughly 20% of weekly hours of work. However, we do notfind significant effects on the labor supply of recipients’ partners.

Most of the large literature related to CCTs focuses on impacts on education and health (seeFiszbein and Schady 2009 for an extensive review). Few studies examine the effects of CCTson labor supply (Skoufias and Di Maro 2006, Maluccio 2008, Foguel and Paes de Barros 2010,Alzúa, Cruces and Ripani 2010). A common feature of these studies is that they exploit theexperimental design of the programs to estimate the causal effect of cash transfers on laborsupply. Thus, our investigation adds to the growing literature of CCTs because, as far as weknow, this is the first paper that identifies the causal effect on labor supply of a CCT programwhich lacks an experimental design.

The rest of the paper proceeds as follows. We review the related literature in section 2. Insection 3, we describe the characteristics of Juntos. We discuss our identification strategy insection 4. In section 5, we describe the data. Results are presented in section 6 and then wepresent robustness checks in section 7. Concluding remarks follow.

2 Literature Review

Research on labor supply responses to welfare programs has long been a subject of interest foreconomists, especially in developed economies where the expansion of benefit transfer programsto low-income population was initiated during 1960s. Since then, researchers and policy-makershave been concerned on how welfare programs affect working incentives of beneficiaries as wellas the indirect (unintended) effects these transfers may generate on non-targeted populationsliving in localities associated with program deployment. For instance, the effect of welfareprograms such as Aid to Families with Dependent Children (AFDC), the Earned Income TaxCredit (EITC), and more recently the Food Stamp Program in the US along with the WorkingFamilies Tax Credit in the UK has intensively been evaluated (see Moffitt 2002 for an extendedreview and discussion). The discussion of how welfare participation affects labor supply of adults

2

can be divided according to (i) the predicted effects of the static and dynamic theoretical modelsof individual labor supply, (ii) program conditions, and (iii) models of household labor supply.

2.1 Theoretical Considerations

The potential effects of benefit transfers can be explained based on the basic static model of laborsupply, which claims that individuals maximize between consumption and leisure (assuming thatleisure is a normal good) facing a budget constraint, which is merely composed by labor (wage)and non labor (initial wealth and monetary or in-kind transfers) income.

In this study, for simplicity, we focus on the role CCTs can play in determining workingincentives1. As pointed out by Alzúa, Cruces and Ripani (2010), CCTs have four potentialchannel through which adult labor supply could be affected.

First, cash transfers represent an increment in non-labor income. Given that no conditionsare imposed with regard to labor effort of beneficiaries, this lump-sum transfer is a pure incomeeffect, and therefore, both employment and working hours are expected to fall. Second, pro-gram conditions can also alter working schedule of adults. For instance, most of the conditionsattached to cash transfer programs imply school enrolment and a maximum number of daysaccepted for children to be absent from school. This increase in school attendance of children al-lows parents to augment labor participation and working hours as well, for they avoid allocatingtime to childcare.

However, school attendance of children might also reduce household’s labor income if childlabor is crucial in determining total family income. This constitutes the third channel throughwhich adult labor supply could be affected. Finally, the fourth path is associated with indirecteffects in the local economies regarding the program deployment. Using a sample from MexicanProgresa welfare program, Angelucci and De Giorgi (2009) find positive effects of cash transfers

1In-kind transfers can affect working behaviour in a very different way than monetary transfers do. Contraryto direct cash transfers, in-kind transfers are supposed to increase consumption, and therefore affect labor supplythrough, at least, three possible paths (holding wages and prices unchanged). First, since in-kind transfers areusually attached to a single or a reduced number of goods (say, food or clothing), this reduces the out-of-pocketexpenditures on such consumables. The reduction of these expenditures is tantamount an increase in non laborincome, and given that this is a pure income effect, this would predict a decrease in working effort. Nonetheless,it would depend on the intended beneficiary. For example, if children are benefited from nutritional programsin their schools, then parents are encouraged to send their children to school and therefore can increase theirlabor participation or working hours. Second, in-kind transfer programs are usually tied to working effort. In thisregard, opposite to cash transfers, families with zero working hours or reduced employment are those commonlyeligible for being in-kind transfer beneficiaries and benefits are reduced for each earned monetary unit. Thus,families may have incentives to reduce their labor force participation in order to become eligibles or to maintainbenefits invariant, which is associated with a substitution effect. Third, labor force responses to in-kind transferswould also depend on the weight families allocate to the particular good or goods in the spectrum of consumables,and henceforth, in the budget constraint. To illustrate, assuming that health expenditures represent a considerableshare of the household budget constraint, programs aiming to reduce health expenditures (e.g., Medicaid) mightreduce labor force participation of adults who otherwise would have had to work intensively in order to meetthose expenses. Combined with attention required by injured household members (children, for instance), theseprograms would reduce labor force participation. Similarly, if the particular in-kind benefit represents a minorshare of the budget constraint, the predicted effect is supposed to be negligible. For a further discussion of therelationship between in-kind benefits and labor supply, see Currie 1993, Yelowitz 1995, Blundell and MaCurdy1999, Moffit 2002, and Hoynes and Schanzenbach 2009.

3

to beneficiaries on consumption of non beneficiaries in villages where the program was randomlyimplemented. Alternatively, qualitative studies (Segovia 2011, for example) have described theappearance of fairs ever since CCTs appeared in localities2.

Dynamic models of labor supply can also add insights to the predicted working effort re-sponses to cash transfers of beneficiaries. According to these models, individuals can replacecurrent for future working hours (and so does consumption) if they perfectly anticipate cashtransfer dates and their inter-temporal discount rate between adjacent periods is near one (thatis, individuals worth current and future working hours the same so that costs for time allocationacross periods tend to zero).

This would imply that (all else equal) individuals can smooth labor intensity in days prior tothe payment dates if they act in a forward-looking behavior (see Card, Chetty and Weber 2007).However, individuals can also have preferences for cash-on-hand or disposable income in orderto alter labor supply behavior. Through this consideration, current non labor income is not theonly path whereby labor supply can be affected by CCTs, but also inter-temporal preferencesand time allocation in a dynamic framework.

Another important consideration is whether welfare programs impose arbitrary restrictionson adult labor supply in order to circumvent working disincentive. Despite the initial uncondi-tional intent related to working effort, some developed countries have indexed program benefitsaccording to the labor supply behaviour of eligibles. For instance, the Temporary Assistance forNeedy Families (TANF) program in the US (formerly known as the AFDC) initially imposesthat at least 20 percent of TANF recipients in each State participate in work or work-relatedactivities for a minimum of 20 hours per week. These activities include regular employment, sub-sidized employment, commuting, on the job training, and 12 months of vocational training foryoung beneficiaries aiming to participate in the labor force. Alternatively, the EITC program,also in the US, consists of a refundable tax credit for low- and medium-income families whichincreases according to a standard range of annually labor income and the number of qualifyingchildren in the family3. These types of cash transfers, both conditional on minimum workinghours or increasing with earned income, act like a contract rigidity, not allowing individuals tomake optimal allocation of working hours. Thus, especially in the case of the EITC where thebenefit is attached to labor income, the response on individual working effort would depend onwhich of the two possible effects - substitution or income - prevail. Empirical findings suggestthat it is participation (entry) rather than hours of work which responds to the EITC (see Eissaand Hoynes 2006 and references therein).

Contrary to these “tied welfare benefits”, CCTs in Latin American do not restrict eligibilityon labor force participation. This implies that the convexification or looseness of the budgetconstraint due to the welfare benefit introduces a pure income effect, hence, encouraging bene-ficiaries to demand more leisure. Further, if those individuals that are not eligibles, say becauseof being just above the poverty line, reduce their working effort in order to diminish total income

2Other studies suggest that individuals are likely to invest in agricultural related productive assets. In a recentarticle, Duflo, Kremer and Robinson (2011) document that demand for agricultural tools tend to increase in datesnearby payment days or seasons of harvest.

3In order to qualify, children must be 18 years old or under (with few exceptions accepting families withchildren “permanently and totally disabled” aged 19 or above), must be somehow related to the claimant (blood,marriage or law), and must be resident of the United States.

4

and “cheat” the system to become eligibles, then the net effect of the CCTs on labor supply ofbeneficiaries would depend not only on the amount of reduced working hours of the ever-eligiblesand the formerly ineligibles, but also in the behavior of the latter group once they have beenselected as program beneficiaries and the transfer have been received (e.g., they can return totheir initial - optimal - working intensity)4.

Lastly, since cash is usually transferred to a particular household member (i.e., the mother),it is worth taking into consideration how welfare is distributed among family members. Forthis reason, theoretical considerations of models of household labor supply can also incorporateuseful ideas. In this line, aside from the potential effects of CCTs on individual adult laborsupply, there exists an open debate on whether families pool their welfare resources. Accordingto this hypothesis, family members act as if they are maximizing a single utility function. Put itdifferently, there exists consensus about the redistribution of household resources among familymembers which are supposed to behave altruistically to each other. Two separate models havebeen developed associated with this “unitary” behavior: the “agreement” (Samuelson 1956)and the “dominant family member” frameworks (Becker 1981). Maximizing a single utilityfunction implies that, regardless of who receives the welfare income and the program targetedbeneficiaries within the family, each of the family members would benefit from the monetarytransfer because of the intrafamily allocation process. In contrast to this “common will” frame,individual cooperative utility models of intrafamily bargaining processes (Manser and Brown1980, McElroy and Horney 1981, and Lundberg and Pollak 1993) as well as non cooperativebargaining models (Lundberg and Pollak 1994) have also been postulated. In these models,income is administered by a single agent within the family (for example, the mother) and thusallocation of resources on consumption and leisure could differ across household members.

Recent empirical evidence based on reduced form estimates instead of structural models,however, indicates that single cooperative utility functions prevail in the family bargainingprocess. Regarding welfare benefits, Lundberg, Pollak and Wales (1997) test the hypothesisof whether families pool their resources exploiting a UK policy change which dictated that childallowances were to be transferred exclusively to wives (mothers). The authors find evidence thatthis policy change induced women to spend more resources on women’s and children’s clothingrelative to men’s clothing. Likewise, Duflo (2003) finds that when pensions to the elderly inSouth Africa are received by women (grandmothers) instead of men, the physical health of girls(granddaughters) tend to improve relative to that of boys living in the same household, whichimplies that resources are reallocated favoring human capital formation of girls.

Regarding labor supply, Bertrand, Mullainathan and Miller (2003) suggest that drops inprime-age men’s labor supply are stronger than that of prime-age women when the South Africanpension benefit is received by women5. In a recent study, Ardington, Case and Hosegood (2009)discuss that pension benefits could, in the case of perfect resource sharing within the family,reduce hours of work and participation of adults, or in the case of imperfect credit markets,social pensions can be used as a credit support for job seekers.

4See Moffitt (2002) for a further examination of this particular scenario.5These results, nevertheless, have been questioned by Posel, Fairbun and Lund (2006), arguing that it is

household resident members who reduce labor force participation. Instead, household receiving social pensions(i.e., those who have at least one men aged 65 or over or one women aged 60 or over) are more likely to havemembers who have migrated to work or looking for work outside the locality.

5

2.2 Previous Empirical Findings in Latin American Countries

To the best of our knowledge, five empirical studies have been carried out addressing the potentialeffects of CCTs on adult labor supply in Latin American Countries. Identification strategies ofall of these studies are based on the random nature of the treatment (most of them at the villagelevel) of the CCTs across the targeted population.

Parker and Skoufias (2000) exploit the experimental design of Mexican Progresa program(currently known as Oportunidades), which randomly assigned treated and control villages, toaddress the question of whether CCTs alter labor participation and overall leisure time of adults,finding no significant effects of Mexican Progresa program on participation rates in the laborforce. Instead, they find that women are more likely to reduce hours allocated to leisure mainlybecause of program commitments such as taking children to schools, clinics and participating incommunity work.

In a later study, Skoufias and di Maro (2006) evaluate the effects of Progresa on outcomesmeasuring adult labor supply. Alike Parker and Skoufias (2000) their identification strategyrelies on a difference-in-difference estimation procedure comparing eligible adults living in treatedvillages (beneficiaries) versus eligible adults living in non treated low-income Mexican villages.The authors do not find a statistically significant effect of CCTs on the probability of beingoccupied. Moreover, based on the fact of random assignment of the program across villages,the authors find that cross-sectional estimates of CCTs on working hours of adults living intreated villages are not statistically different from working hours of adults residing in (randomly)untreated villages. Using a similar estimation methodology for Nicaraguan Red de ProtecciónSocial (RPS) program but analyzing the overall household labor supply, that is, the sum of eachmember’s labor intensity, Maluccio (2008) finds a negative small but statistically significanteffect of the program on household hours of work, especially in agricultural activities. Theauthor argues that this reduction is explained based on the fact that these activities are perhapsassociated with lower marginal rates of return. In contrast, Foguel and Paes de Barros (2010) findno statistically significant effects of six Brazilian programs (Bolsa Escola, Bolsa Alimentacão,Bolsa Familia, among others) on adult labor supply, neither on the extensive nor the intensivemargin.

Finally, Alzúa, Cruces and Ripani (2010) find negative but small -if not inexistent- effects ofthree different programs from Latin American countries (RPS in Nicaragua, Progresa in Mexico,and Programa de Asignación Familiar -PRAF- in Honduras) on labor force participation andthe probability of migrating from agricultural to other working activities. However, they dofind a reduction of about 4.7 to 6.3 weekly hours worked in the case of Nicaraguan RPS and apositive and significant effect of Mexican Progresa program on male wages.

These studies rely on the experimental design of the different programs evaluated, and mostof them (with the exception of Skoufias and di Maro 2006) fail to control for the possibilityof reallocation of working effort of ineligibles in communities or villages regarding programdeployment, as pinpointed by Angelucci and De Giorgi (2009). Not taking into account thispotential effect may introduce negative bias (in absolute terms) to the parameters of interestassuming that ineligibles are more prone to increase their labor intensity given the increase inthe demand for consumable goods and agricultural productive assets in days nearby the transferschedules. Because this potential increase in the demand of a particular set of goods may increase

6

real wages of ineligibles (hence introducing a substitution effect), previous empirical findingsbased on double-difference comparisons are likely to understate the labor supply responses toCCTs.

In the following lines, we attempt to add empirical evidence of the effect of CCTs on adultlabor supply based on the Peruvian Juntos program. In contrast with the previous empiricalstudies, Juntos was not originally experimentally designed, so the identification strategy is basedon a sort of “natural experiment” taking advantage of the difference between transfer and inter-view dates. Albeit this variation seems to be exogenous, plausible and testable assumptions areneeded in order to consistently estimate adult labor supply responses to welfare transfers.

3 The Program and Its Mechanics

Following its Latin American counterparts, the Peruvian government launched a nation-wideCCT program in 2005. The program, named Juntos, seeks to reduce current and future povertythrough cash transfers and investments in the human capital of children. Initially, Juntos wasimplemented in 70 municipalities with a budget of US$ 45 million. In 2009, 409,000 familieswere direct beneficiaries in 638 municipalities and the budget was raised to US$ 260 million.The amount of the transfer is 100 Peruvian Nuevos Soles (local currency) every month, whichis equivalent to 12% of the monthly household expenditure in our sample 6. Once the family isenrolled in the program, transfers are given to the female head of the household according to apayment schedule defined by Juntos7.

It is worth noting that payment dates are defined at the village level which implies thatsome municipalities have more than one payment date. Juntos sets a particular day in everyvillage so we have some within-municipality variation in payment dates. However the maximumdifference between the earliest and the latest payment date is smaller than 7 days. This featureof the program does not represent a major problem to our strategy as it will be shown in section7.

How do beneficiaries receive the cash transfer? In 2009 there were two mechanisms. Themain way to receive the cash was to go to the local branch of the Peruvian National Bank andwithdraw the money (54% of the beneficiaries in our sample). The second way was to go to themain square of the village on the day of payment and wait for an armored van which containedthe money. One difference between these methods is that the former allows the beneficiary togo to the bank at some other day while the latter does not. Both systems are mutually exclusiveat the village level so beneficiaries do not choose the way they get the money. We discuss theimplications of these mechanisms in the next section.

The program does not impose any constraint on the use of the money, however, all beneficia-ries must meet the following conditions: i) children of age 6-14 years attend at least 85% schoolclasses; ii) children of age 0-60 months get fully immunized and visit health centres where theirgrowth is measured and vitamins are provided; iii) children of age 3-36 months get nutritionsupplements; iv) pregnant women visit health clinics for prenatal care; v) lactating women visithealth centres for post-natal care; vi) parents attend health clinics to receive information about

6Alternatively, the payment is equivalent to 63% of monthly per capita expenditure.7Since 2010, the cash transfer is made every two months.

7

nutrition, health and hygiene; vii) parents without ID (identification) attend the program Minombre (My Name).

The conditions outlined above are very similar to those of other programs such as Oportu-nidades in Mexico. Given that conditions are related to investments in education and health,these kind of outcomes have received more attention than others. More specifically, Perova andVakis (2011), using IV and matching methods, find that Juntos has increased consumption andschool enrolment. Sanchez and Jaramillo (2012) show that the program has reduced early mal-nutrition among children in treated households. However, it is still relevant for policy-makersto assess whether Juntos has impacts on the labor supply of its beneficiaries. Now, we turn todiscuss our identification strategy.

4 Identification Strategy

Previous studies (Skoufias and di Maro, 2006; Maluccio, 2008; Alzúa et al., 2010) have relied oncomparisons between beneficiaries and non-beneficiaries to identify the impact of CCTs on laborsupply. Those estimates may be unbiased when randomization is possible but many programslack an experimental design. In this paper, we propose an alternative strategy, which exploitsdifferences between interview dates of the National Household Survey (Encuesta Nacional deHogares - ENAHO) and payment dates of Juntos.

In particular, we will explore whether labor supply is reduced in the days near the paymentdate. To do so, we compare beneficiaries, within the same municipality, who were interviewedjust after the payment to those who were not. Given that most households members are engagedin agricultural and highly-flexible occupations (i.e. self-employed), individuals may decide towork less in the week following the payment date.

Though we exploit within-municipality variation in interview dates of ENAHO, our measureof distance is constructed as the difference between the payment date and the week previous tothe survey. These seven days prior to the interview day are called the “reference week”. Wheninterviewers survey households, they usually ask household members whether they have donespecific activities during the previous seven days. For example, when asking about labor forceparticipation, interviewers ask the following question: “during the last week, from [day 1] to[day 7], did you have any job?”. Thus, our dependent variable is the hours of work during the“reference week”.

To illustrate, Figure 1 plots hours worked in the reference week for distinct groups of bene-ficiaries according to the distance (in weeks) between the payment date and the reference week.The decline of hours worked during the reference week is linked to the week in which the trans-fer is received for all individuals included in our sample. Nonetheless, this decline is larger forrecipients of cash compared to their partners. The largest decrease in working hours happenswhen the payment occurs one week before the reference week and it returns to its original levelwhen the payment has not been done yet (the transfer would occur at least one week after thereference week).

Variation in payment and interview dates is crucial to our strategy. In Table 1, we presentthe distribution of payment dates associated with the cash transfer from Juntos. Regarding theday of the month, we do not find any special pattern. If anything, we could say that there is a

8

slight concentration around the third week of the month, between the 16th and the 20th day.Regarding the day of the week, it seems that Mondays are the most common day of paymentwhile Sundays are the least frequent. The distribution of interview dates is presented in thebottom half of the table. If we look at the day of the month, the frequency of dates looks prettybalanced. We also note that almost all interviews are conducted on Sundays, when most of thefamily members stay at home.

For the empirical analysis, we construct four dummies according to the distance betweenthe payment date and the reference week. Specifically, the first is equal to one if the paymentdate takes place at least two weeks before the reference week. Similarly, the second dummy isequal to one if the payment is made within the week prior to the reference week. The thirdvariable takes the value of one when the payment from Juntos occurs at some point during thereference week. The last dummy denotes that payment takes place after the reference week.Each dummy may capture a specific effect related to the distance between the date of paymentand the reference week.

For instance, the second dummy could capture the time spent (during the reference week) onpurchasing goods with the cash received. Similarly, the third dummy may capture the reductionin hours of work related to the time that the cash’s recipient needs to go to the bank andwithdraw the money. Also, the fourth may capture a “anticipation” bias from beneficiaries.

Given that the distance between date of payment and date of interview (reference week) isexogenous, our empirical equation is :

yij = λj +∑

k

δkdij +X′iβ + µij (1)

where yij is the outcome variable (participation, hours of work), λj is a municipality fixed effect,dij denotes a specific distance (in days) between date of payment and date of interview, Xi is avector of covariates such as age, education, native language and so on, and µ is the error term.In the following analysis, the omitted category is that the payment was done at least two weeksbefore the reference week.

There are two potential threats to the validity of our strategy. On the one hand, it maybe possible that when the interviewers of the ENAHO arrive at a given municipality, they gofirst to families who work less and later to families who work harder. If this were the case,our estimates should be seen as a lower bound (in absolute terms)8. On the other hand, ourindicator variables may capture other effects not related to the transfer but correlated with otherunobservable variables. To check that this is not the case, we conduct a falsification test onlyincluding non-beneficiaries in the sample. The details of this procedure would be presented insection 7.

Finally, two limitations of the data may affect our four dummies of interest. First, we onlyhave information about payment dates established by Juntos but we fail to observe the actualdate the beneficiary went to the bank and withdrew the money9. Second, in some municipalities,there may be two or more payment dates. For example, in a given municipality, there could

8This is because our coefficients are calculated as a function of the omitted category, which is, those who werepaid more than one week before the cash transfer (i.e., those who were interviewed first).

9This would be true only in the villages where the payment method is through the bank but not in the villageswhere beneficiaries go to the main square on the payment day to wait for the armored van.

9

be two villages and each of them may have a different payment date. However, the ENAHOonly provides information up to the municipality level. Thus, we are unable to identify whichhouseholds live in, say, village 1 and who lives in village 2. In these cases, we define themunicipality payment date as the first date (the earliest) of payments.10.

5 Data, Variables and Descriptive Statistics

5.1 Data

Our primary source of information is the ENAHO conducted in 2009 by the Instituto Nacionalde Estadística e Informática (INEI). The ENAHO 2009 collects individual level information andis a nationwide representative survey, both in urban and rural areas. We use information fromthe employment and income registry, which restricts the sample only for individuals aged 14or older. The ENAHO has three important features. First, it includes several questions whichallow us to accurately indentify households receiving monetary transfers from Juntos. This isparticularly important since the program design refers to women as the only household’s transferrecipients. Second, this survey includes questions regarding relationship with the family head,enabling us to distinguish the potential impact for different household members, say male headsand female spouses (or, equivalently, cash’s recipients). Finally, this dataset provides a rich setof variables that allows us to construct different labor supply outcomes and include a wide setof controls in our regressions.

To precisely estimate the impact of the proximity to the payment date on labor supplyoutcomes we need a representative sample of all municipalities which are beneficiaries fromJuntos. By 2009, 638 municipalities were part of the program. Given that the ENAHO followsa stratified sampling procedure, this survey collected information in 260 municipalities enrolledin Juntos in this particular year. This represents roughly 40.8% of the municipalities in whichthe Juntos program was present in 2009.

Nevertheless, when expanding the sample using the survey weights from the sampling de-sign, Perova and Vakis (2011) find that the number of households which report receiving cashtransfers from Juntos surveyed in the ENAHO 2009 is very close to the number of beneficiaryhouseholds listed in the official registries. We therefore use sample weights in all of our regres-sions and correct standard errors based on sampling design. This procedure should guaranteethat estimates arising from our sample are representative average effects of the proximity to thepayment date on labor supply incentives for all the program beneficiaries.

As an additional concern we check whether the transfer conditions were consistently re-produced in each of the surveyed households. In other words, we check that (i) the householdtransfer receptor is the mother (female head or household head’s spouse), (ii) the monetary trans-fer reported by the woman is equivalent to 100 Peruvian Nuevos Soles (about 37 US currentdollars), and (iii) the frequency of transfers is monthly. Around 98% of the transfer recipientsin our sample were women satisfying the mentioned conditions.

Further, we check that the surveyed households which reported having received monetarytransfers from Juntos satisfy the eligibility conditions. Regardless of the fact that eligible house-

10In section 7, we use the last date of payment and our results remain unchanged.

10

holds should be below the poverty line in order to receive the transfer, our working samplesuggests that about 19% of the households were above the poverty line defined by INEI 11.

Unique municipality identifiers are used to match the information of payment dates from theadministrative dataset, previously collapsed at the municipality level, to the beneficiaries samplebuilt up from the ENAHO 2009. Our final sample contains information of 3,781 individuals livingin 1,215 households enrolled in Juntos.

5.2 Outcome Variables

We focus on three different measures of labor supply behavior: participation (extensive margin)weekly hours worked (intensive margin) and working for paid activities. As described above,each of the outcome variables are defined for the week before the day of the interview (whichusually takes place on Sundays). Labor participation is a dummy variable equals to one whenthe individual reported having worked or searching for a job any time during the seven daysprior to the interview. To measure labor intensity, we take the total number of hours workedduring the same week. These two variables are commonly used in empirical studies (Skoufiasand di Maro, 2006; Maluccio, 2008; Fogel and Paes de Barros, 2010; Alzúa et al., 2010) andhenceforth are also useful to make comparisons of adult labor supply responses to cash transfersacross Latin American countries. Lastly, the indicator for working for paid activities is relevantfor evidencing changes in labor supply alternative margins once the payment has already beendone or is about to occur (for instance, household members could reallocate time to family orhome production related unpaid activities once the cash has been transferred).

Given that we have information of the number of hours worked in each day of the referenceweek, we are able to test whether individuals change their labor supply behavior in a given dayor whether they balance their labor intensity throughout the whole week. This insight will behelpful when interpreting our main results.

5.3 Descriptive Statistics

Variable averages and standard errors (reported in parentheses) are shown in Table 2. Eachcolumn reports summary statistics of all individuals included in each of our four dummiesof distance. The average individual is about 42 years old. With regards to the educationalattainment, 18% of individuals report not having reached any regular basic educational level,60% have (incomplete) primary level education, and 17% reported at least one year of secondaryeducation. About 1% of individuals have at least one formal year of tertiary education and only1% of individuals have completed tertiary education. We also include native language indicatorsin order to capture race heterogeneity. Individuals are 65% likely to report Quechua as theirnative language. Both the high percentage of Quechua speakers and the 85% of individualsliving in rural areas suggest that our sample is mainly composed of indigenous people. Giventhe possibility of filters among the program regarding non-poor people receiving cash transfers,

11The reason underlying the filters of non poor households as part of the Juntos beneficiaries can be explainedbased on poverty transitions (households being initially poor and then escaping from poverty once they had alreadybeen selected as beneficiaries) and program administrative failures (non poor households selected as beneficiarieseven when the program was initially targeted to households below the poverty line).

11

we include three poverty indicators: (i) non-poor, (ii) poor, and (iii) extremely poor. Theseindicators are calculated by INEI using a poverty line and the monthly per capita expenditureat the household level as a welfare measure. Around 21% of individuals in the whole sample arenon-poor, while roughly 42% are extremely poor and 37% are poor. The bottom part of thetable shows descriptive statistics of the outcome variables used in the analysis. Around 94% ofindividuals reported participating in the labor force. The number of hours worked is about 30per week. Finally, 56% of individuals in our sample reported having worked for paid activities.

Table 2 also shows some heterogeneity between groups. However, these differences in char-acteristics such as education and native language seem negligible. In the following empiricalanalysis we include this set of variables and municipality fixed effects in order to control forthese slight differences. In section 6 we present the results arising from the transfer proximitymodel described in the previous section.

Main regressions are estimated only for poor individuals to circumvent potential problemsassociated with the possibility of self-selection which can bias our results (e.g., individuals whoare not poor can mislead the program targeting procedure in order to become eligibles). However,we also present evidence that when including non-poor individuals in our regressions, coefficientsdo not significantly differ.

6 Results

6.1 Main Results

Table 3 reports the resulting estimates for the equation of labor force participation. Each rowindicates the distance between the cash transfer and the reference week. Columns (3), (6) and (9)are our preferred specifications since they include municipality fixed effects as well as individualcovariates. Results from these columns suggest that there are no effects on the extensive marginof the labor supply (i.e. participation) even when splitting the sample by recipient and recipient’spartner.

Table 4 shows results for the equation of hours worked in the reference week. For thesample as a whole, there are no significant effects on the intensive margin. However, we findthat having received the cash transfer within the seven days before the reference week reducesabout 5.7 hours of work in the reference week for recipients only - see column (6). Recallthat the effect of the transfer among recipients may be driven by three possible confoundingfactors: (i) anticipation bias (increasing demand for leisure just before the transfer is made);(ii) time spent in transportation from the location of residence to the bank; and (iii) time spentin purchasing the goods or consuming the money once it has been withdrawn from the bank.Under the assumption that those who were paid during the reference week have also anticipatedthe transfer date (and, therefore, have reduced their working hours) and have spent some timein receiving the transfer, then the resulting point estimate for those who were paid the weekbefore the beginning of the reference week is not driven by these particular confounding effects.Nonetheless, time spent in purchasing goods with the received money could also be affecting ourestimates12.

12It is worth noting, however, that the reduction in working hours occurs in the reference week. So if there

12

Now, we explore if the distance between the cash transfer and the reference week affects thedecision of working for paid activities. The dependent variable is a dummy which is equal to oneif the individual is engaged in a paid-job and is zero, otherwise. Results from this estimationare presented in Table 5. The estimated effects are insignificant but in the case of recipients,they are negative while in the case of their partners the estimates are positive. These resultsmay reflect some rigidities to switch from unpaid jobs to paid jobs in the short run.

In addition, we test whether the reduction in hours devoted to working activities is concen-trated in a particular day of the reference week. Under the hypothesis that the reduction inhours of work is being driven by time spent in purchasing goods (once we control for the poten-tial anticipation and transportation effects), one should expect that the effect of the transfer isgrouped in a particular day of the week (say, the day which is closer to the payment date). InTable 6 we report the resulting coefficients for every day of the reference week. Consistent withthe estimates shown in Table 4, we find negative and significant effects for those who are paidwithin the seven days before the reference week. Specifically, we find that working hours reduceby roughly 1.3 hours in every day except for Sundays. In addition, we find that hours of workon Thursday reduce by 1 hour if payment occurs in the reference week.

These results show a decrease in hours worked when payment occurs in the reference week.This reduction is most likely to be driven by time spent on going to the bank. However, whenpayment takes place one week before the reference week, the reduction in labor intensity is evenlydistributed along reference week which is inconsistent with the hypothesis that our results aremainly driven by transportation from the household to the bank. Thus, the dummy “during thereference week” captures the reduction driven by transportation while the dummy “one weekbefore” reflects the disincentives to work generated by the having received the cash transfer.

6.2 Heterogeneous Effects

The purpose of this section is to explore whether there are heterogeneous effects of the cashtransfer on weekly hours of work 13. We begin by splitting the sample according to married andnot-married women. Results from these estimations are shown in Table 7. Interestingly, we findthat if cash transfer is made in the week prior to the reference week, the reduction in hours ofwork is larger - 11 hours- for married women than for not-married. One potential explanationfor this difference is that married women also rely on their husbands’ income and this allowthem to reduce their labor supply more than not-married women.

Next, we analyze if there is heterogeneity between young and old recipients. In order tokeep a balanced sample in both groups, we say that a recipient is young if she is 40 years oldor younger and she is old, otherwise. Table 8 presents results from this specification. Thepoint estimate of the effect of being paid one week before the reference week is larger for youngrecipients - 12 hours- than for old recipients. From a theoretical point of view, this evidence isconsistent with young people having younger children or higher discount rates14 and therefore,

exists an effect encompassing time spent in consumption of goods, then it is likely that this effect should appearjust after the transfer has been done, but not in the reference week (seven days after the payment date).

13We should mention that when comparing independent v.s. dependent beneficiaries and highly-educated v.s.low-educated recipients , we find no difference (results not reported) in the effects.

14Behavioral economists suggest that young people is usually more present-biased than old people.

13

reacting more than old people, to the same income shock.Lastly, we distinguish between recipients who have children aged 5 or less and those who do

not. This distinction is important because the presence of young children at home is a majordeterminant in female labor supply. Results from splitting the sample according to children’sage are presented in Table 9. As expected, recipients with children aged 5 or less reduce theirlabor supply than recipients with older children. The point estimate of having received the cashtransfer one week before the reference week is -9.96 hours for recipients with young children.This could suggest that recipients reduce their hours of work in order to spend this additionaltime taking care of their children.

Taken together, these heterogeneous effects provide evidence which is consistent with an non-labor income shock. They also shed some light on what mechanisms explain our main results.Now, we turn to question some assumptions made and perform some robustness checks to seewhat happens if these assumptions do not hold.

7 Robustness Analysis

When we identified the beneficiaries from Juntos, we excluded non-poor households. However,in our data some of them claimed that they were receiving the cash transfer from the programon a monthly basis. Although Juntos is targeted at the poor, it is possible that households whowere poor when Juntos arrived at their municipalities escaped from poverty along the years.In Table 10 we present estimates from equation (1) but including non-poor beneficiaries (onlyrecipients). We find that having been paid one week before the reference week reduces by 6hours the labor supply of beneficiaries -see column (3). These estimated effects are slightlylarger than those presented in Table 4. This additional evidence suggests that the reductionin hours of work is not driven by the time needed to withdraw the money given that non-poorbeneficiaries are more likely to spend less time going from home to the bank. Moreover, thisdifference may suggest that the labor supply of non-poor beneficiaries is more elastic than thatof poor beneficiaries.

A major threat to our identification strategy is that the dummies of distance between pay-ment dates and interview dates may be capturing other variables not related to the cash transfer,but to the specific date of the payment. For instance, it could be that payment dates are es-tablished on days when the labor supply is low for a different reason than the transfer (e.g.holidays). This potential correlation between dates and unobservable variables that affect hoursof work would invalidate our strategy. To check that this is not the case, we perform a falsifi-cation test using data from non-beneficiaries. If our dummies are correlated with variables thataffect labor supply, they should also have an impact on the hours of work of non-beneficiaries.Thus, we estimate equation (1) but only including non-beneficiaries in our sample15. Table11 presents results from this estimation. Not surprisingly, we find that none of our distancedummies are significant at any conventional level. This evidence suggests that our indicatorvariables are not correlated with omitted variables that may affect the labor supply. Based onthese results, our identification strategy does not seem to be invalid.

15We include spouses of household heads who did not report to be beneficiaries from Juntos but who lived inmunicipalities that the program has reached.

14

Recall - see section 3- that some municipalities may have more than two payment dates. Inthe previous analysis we have used the first payment date. Now, we re-estimate equation (1)with the set of four dummies but using the last payment date instead of the first. In Table 12,we present results from this estimation. The first three columns report results for only poorrecipients and the other three include non-poor recipients in the sample. In column (3) we findthat the effect of having being paid one week before is slightly smaller than when we used thefirst date of payment (column (6) in Table 4) but it is still highly significant. In column (6),we include non-poor recipients in the sample and the estimated effect is larger than in column(3). Also, it is remarkably similar to that of column (3) in Table 10 (when we used the firstdate of payment). Thus, the impact of having received the cash transfer one week before thereference week on hours of work does not significantly change when we modify the definition ofthe municipality-payment date.

Even after all these checks one may still argue that our results are being driven by thetime recipients spend in going to the bank. In order to rule out this possibility, we split thesample according to the payment mechanism defined by Juntos at the village level. The paymentmechanisms are: i) going to the bank (which may not be at your village of residence) at thedate defined by Juntos or later and ii) going to the main square of your village to wait for anarmored van and receive the cash on the same day defined by Juntos. In Table 13, we presentthe estimates of these regressions. As we can see, the estimates are not significant for thosebeneficiaries who went to the bank. In contrast, the effects of receiving the cash transfer oneweek before is large - about 9 hours- and highly significant for recipients who went to the mainsquare of their village. This last piece of evidence tells us that the reduction in hours of work isdue to an income shock and is not related to the time the beneficiary needs to go to the bank.

8 Concluding Remarks

It is well-known that welfare programs in developed countries have unintended effects on laborsupply (Moffit 2002). In spite of this evidence, there have been few efforts to identify the impactsof CCTs on the labor supply of their beneficiaries.

In this paper we make a first attempt to estimate the effects of cash transfers on labor supplywithout using an experimental research design. Given the high flexibility of rural occupations(mostly agricultural), our approach consists of studying the behavior of beneficiaries in daysnear to the payment dates of Juntos. In particular, we find that having been paid one weekbefore the reference week reduces the labor supply of female heads (cash recipients) by about 6hours in the mentioned week.

Some interesting policy implications arise from our findings. First, CCTs could have largereffects on education and health if mothers that work less are encouraged to invest more timewith their children. Second, changes in the frequency of payments may alter the magnitude ofthe estimated effects in this study. Third, it should be analyzed whether it is feasible to offertraining or new technologies for agricultural activities (e.g. use of fertilizers) near the paymentdates. Based on results from Duflo et al. (2011), these special offers could have large impactson productivity given that households have extra time and money during these days.

Finally, we believe that our strategy could be used to analyze other interesting outcomes as

15

well. For instance, we could examine if households change their regular consumption patternduring the week after the payment (e.g. going to restaurants instead of eating at home). Also,it would be relevant to see if recipients of the cash do not lose control over the money onceshe arrives home. If potential disputes within the household arise after the payment, we couldtest whether there is an increase in domestic violence during these days. These are promisingavenues for future research that may expand the discussion about the benefits and limitationsof CCTs.

16

ReferencesAlzúa, M. ; Cruces, G. ; Ripani, L. (2010): “Welfare Programs and Labor Supply inDeveloping Countries. Experimental Evidence from Latin America”; CEDLAS Working Paper0095, January 2010

Angelucci, M. and De Giorgi, G. (2009): “Indirect Effects of an Aid Program: How DoCash Transfers Affect Ineligibles’ Consumption?”; American Economic Review, 99 (1), pp.486-508

Becker, G. (1981): “A Treatise on the Family”. Cambridge: Harvard University Press

Bertrand, M.; Mullainathan, S. and Miller, D. (2003): “Public Policy and ExtendedFamilies: Evidence from Pensions in South Africa”; World Bank Economic Review, 17(1), pp.27-50

Blundell, R. and Macurdy, T. (1999): “Labor Supply: A Review of Alternative Ap-proaches”; Handbook of Labor Economics, V. 3, Edited by O. Ashenfelter and D. Card; Elsevier

Card, D.; Chetty, R. and Weber, A. (2007) : “Cash-On-Hand and Competing Modelsof Intertemporal Behavior: New Evidence from the Labor Market”; Quarterly Journal ofEconomics, 122(4), pp. 1511-1560

Currie, J. (1993): “Welfare and the Well-Being of Children: The Relative Effectiveness ofCash and In-kind Transfers”; NBER Working Paper No. 4539

Duflo, E. (2003): “Grandmothers and Granddaughters: Old-Age Pensions and Intra-household Allocation in South Africa”; World Bank Economic Review, 17(1), pp. 1-25

Duflo, E. ; Kremer, M. ; Robinson, J. (2011): “Nudging Farmers to Use Fertilizer: Theoryand Experimental Evidence from Kenya”; American Economic Review, 101 (October 2011),pp. 2350-2390

Edmonds, E. and Schady, N.: “Poverty Alleviation and Child Labor”; American EconomicJournal: Economic Policy, forthcoming

Eissa, N. and Hoynes, H.(2006): “The Hours of Work Response of Married Couples: Taxesand the Earned Income Tax Credit”; in Tax Policy and Labor Market Performance, Jonas A.and Birch Sorensen, P. eds. MIT Press, 2006

Fiszbein, A. and Schady, N. (2009): “Conditional Cash Transfers: Reducing Present andFuture Poverty”; World Bank Report, Washington 2009

Foguel, M. and Paes de Barros, R. (2010): “The Effects of Conditional Cash TransferProgrammes on Adult Labour Supply: An Empirical Analysis Using a Time-Series-Cross-Section Sample of Brazilian Municipalities”; Estudos Economicos. Sao Paulo, V.40, No. 2,pp.259-293

17

Hoynes, H. and Schanzenbach, D. (2009): “Consumption Responses to In-Kind Transfers:Evidence from the Introduction of the Food Stamp Program”; American Economic Journal:Applied Economics, 1(4), pp. 109-139

Lundberg, S. and Pollak, R. (1993): “Separate Spheres Bargaining and the Marriage Mar-ket”; Journal of Political Economy, 101(6), pp. 988-1010

Lundberg, S. and Pollak, R. (1994): “Noncooperative Bargaining Models of Marriage”;American Economic Review - Papers Proceedings, 84(2), pp. 132-137

Lundberg, S. and Pollak, R. (1997): “Do Husbands and Wives Pool Their Resources?Evidence from the United Kingdom Child Benefit”; Journal of Human Resources, 32(3), pp.463-480

McElroy, M. and Horney, M. (1981): “Nash Bargained Household Decisions”; InternationalEconomic Review, 22(2), pp. 333-350

Maluccio, J. (2008): “The Impact of Conditional Cash Transfers in Nicaragua on Consump-tion, Productive Investments and Labor Allocation”; The Journal of Development Studies

Manser, M. and Brown, M. (1980): “Marriage and Household Decision Making: A BargainingAnalysis”; International Economic Review, 21(1), pp. 31-44

Moffit, R. (2002): “Welfare Programs and Labor Supply”; Handbook of Public Economics, V.4, Edited by A. Auerbach and M. Feldstein; Elsevier

Parker, S. and Skoufias, E. (2000): “The impact of Progresa on work, leisure, and timeallocation”; Final report, International Food Policy Research Institute, Washington

Perova, E. and Vakis, R.(2011): “Duración e Impacto del programa JUNTOS en el Perú”;mimeo.

Posel, D. ; Fairburn, J. and Lund, F. (2006): “Labour Migration and Households: AReconsideration of the Effects of the Social Pension on Labour Supply in South Africa”;Economic Modelling, 23, pp. 836-853

Samuelson, P. (1956): “Social Indifference Curves”; Quarterly Journal of Economics, 70, pp.1-22

Sanchez, A. and Jaramillo, M. (2012): “Impacto del Programa Juntos sobre Nutriciontemprana”; Peruvian Central Bank Working Papers, DT-2012-01

Skoufias, E. and Di Maro, V. (2006): “Conditional Cash Transfers, Adult Work Incentivesand Poverty”; World Bank Policy Research Working Paper 3973, August 2006

Yellowitz, A. (1995): “The Medicaid Notch, Labor Supply, and Welfare Participation: Evi-dence from Eligibility Expansions”; Quarterly Journal of Economics, 105, pp. 909-940

18

Figure 1: Weekly Hours of Work according to distance from Payment

19

Table 1: Distribution of Payment and Interview DatesPanel A: Dates of Payment Frequency PercentageDay of the month1-5 323 8,56-10 485 12,811-15 726 19,216-20 1006 26,621-25 712 18,826-31 529 14,0Day of the weekSunday 178 4,7Monday 1,099 29,1Tuesday 512 13,5Wednesday 490 13,0Thursday 648 17,1Friday 449 11,9Saturday 405 10,7

Panel B: Dates of Interview Frequency PercentageDay of the month1-5 664 17,06-10 579 14,811-15 892 22,916-20 556 14,321-25 624 16,026-31 584 15,0Day of the weekSunday 3,780 99,97Monday 1 0,03

Sources: Juntos Administrative data (payment dates) and ENAHO surveys (interview dates)

20

Table2:

Descriptiv

eStatist

ics

Distancebe

tweencash

tran

sfer

date

andthereferenceweek

Variable

Atleasttw

oweeks

before

One

weekbe

fore

Duringthereferenceweek

Atleaston

eweekafter

Age

43.53

41.70

42.53

42.27

(12.16)

(10.60)

(12.29)

(11.68)

Male

0.47

0.48

0.46

0.46

(0.50)

(0.50)

(0.50)

(0.50)

Edu

cation

level:Noeducation

0.19

0.18

0.17

0.19

(0.39)

(0.39)

(0.38)

(0.39)

Edu

cation

level:Primary

0.60

0.67

0.64

0.63

(0.49)

(0.47)

(0.48)

(0.48)

Edu

cation

level:Second

ary

0.21

0.14

0.18

0.17

(0.40)

(0.34)

(0.38)

(0.37)

Edu

cation

level:Te

rtiary

0.01

0.01

0.01

0.01

(0.11)

(0.12)

(0.11)

(0.11)

NativeLa

ngua

ge:Sp

anish

0.30

0.28

0.33

0.27

(0.46)

(0.45)

(0.47)

(0.44)

NativeLa

ngua

ge:Quechua

0.61

0.71

0.62

0.68

(0.49)

(0.46)

(0.49)

(0.46)

NativeLa

ngua

ge:Other

0.09

0.01

0.05

0.05

(0.28)

(0.09)

(0.22)

(0.22)

Poverty

status:Extremelypo

or0.40

0.47

0.43

0.47

(0.49)

(0.50)

(0.50)

(0.50)

Poverty

status:Poo

r0.37

0.39

0.36

0.35

(0.48)

(0.49)

(0.48)

(0.48)

Poverty

status:Non

poor

0.22

0.14

0.21

0.18

(0.42)

(0.35)

(0.41)

(0.39)

Livesin

rurala

reas

0.90

0.77

0.83

0.86

(0.30)

(0.42)

(0.37)

(0.35)

Inlabo

rforce

0.95

0.93

0.93

0.95

(0.21)

(0.25)

(0.26)

(0.22)

Weeklyho

ursworked

30.66

29.25

30.19

31.77

(15.68)

(15.91)

(17.97)

(16.22)

Workedforpa

idactivities

0.55

0.58

0.53

0.55

(0.50)

(0.49)

(0.50)

(0.50)

Observation

s563

425

425

565

21

Table3:

Effects

ofdistan

cebe

tweencash

tran

sfer

andthereferenc

eweekon

labo

rforcepa

rticipation

All

Recipients

Recipients’

partner

Tran

sfer

was:

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

One

weekbe

fore

thereferenceweek

-0.028*

-0.009

-0.010

-0.072**

-0.044

-0.043

0.015*

0.006

0.148

(0.016)

(0.030)

(0.030)

(0.029)

(0.060)

(0.065)

(0.009)

(0.007)

(0.203)


-0.027

-0.011

-0.008

-0.054*

-0.046

-0.047

0.002

0.019

0.162

(0.017)

(0.032)

(0.032)

(0.029)

(0.059)

(0.070)

(0.012)

(0.016)

(0.206)

Atleaston

eweekafterthereferenceweek

-0.011

-0.007

-0.011

-0.029

-0.024

-0.035

0.010

-0.001

-0.319

(0.014)

(0.024)

(0.024)

(0.024)

(0.047)

(0.050)

(0.010)

(0.009)

(0.336)

Mun

icipality

fixed

effects

No

Yes

Yes

No

Yes

Yes

No

Yes

Yes

Add

itiona

lcon

trols

No

No

Yes

No

No

Yes

No

No

Yes

Observation

s1,615

1,615

1,615

859

859

859

756

756

756

R-squ

ared

0.003

0.144

0.186

0.008

0.284

0.293

0.005

0.280

0.399

Not

e:C

lust

ered

stan

dard

erro

rsat

the

vill

age

leve

lin

pare

nthe

ses.

Add

itio

nal

cont

rols

incl

ude:

sex,

mar

ital

stat

us,

age,

educ

atio

n,na

tive

lang

uage

indi

cato

rs,

pov

erty

stat

usdu

mm

ies

and

adu

mm

yva

riab

leth

atin

dica

tes

whe

ther

orno

tth

ein

divi

dual

live

sin

aru

ral

area

.F

inal

lyw

ein

tera

ctou

rdu

mm

ies

ofin

tere

stw

ith

adu

mm

yth

atis

equa

lto

one

whe

nth

e

indi

vidu

alis

the

head

ofth

eho

useh

old.

22

Table4:

Effects

ofdistan

cebe

tweencash

tran

sfer

andthereferenc

eweekon

weeklyho

ursof

work

All

Recipients

Recipients’

partner

Tran

sfer

was:

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

One

weekbe

fore

thereferenceweek

-0.558

-1.899

-1.756

-1.640

-4.192*

-5.618**

0.241

-0.425

-0.425

(1.131)

(1.840)

(1.832)

(1.398)

(2.284)

(2.397)

(1.789)

(2.903)

(2.903)


-0.294

-1.234

-0.972

-0.790

-0.594

-1.836

0.196

-1.537

-1.537

(1.164)

(1.849)

(1.835)

(1.414)

(2.213)

(2.367)

(1.872)

(3.017)

(3.017)

Atleaston


1.308

0.678

0.996

0.992

3.802*

2.573

1.614

-3.312

-3.312

(1.073)

(1.825)

(1.806)

(1.300)

(2.206)

(2.295)

(1.729)

(2.918)

(2.918)

Mun

icipality

fixed

effects

No

Yes

Yes

No

Yes

Yes

No

Yes

Yes

Add

itiona

lcon

trols

No

No

Yes

No

No

Yes

No

No

Yes

Observation

s1,577

1,577

1,577

827

827

827

750

750

750

R-squ

ared

0.002

0.259

0.284

0.005

0.395

0.409

0.001

0.419

0.419

Not

e:C

lust

ered

stan

dard

erro

rsat

the

vill

age

leve

lin

pare

nthe

ses.

Add

itio

nal

cont

rols

incl

ude:

sex,

mar

ital

stat

us,

age,

educ

atio

n,na

tive

lang

uage

indi

cato

rs,

pov

erty

stat

usdu

mm

ies

and

adu

mm

yva

riab

leth

atin

dica

tes

whe

ther

orno

tth

ein

divi

dual

live

sin

aru

ral

area

.F

inal

lyw

ein

tera

ctou

rdu

mm

ies

ofin

tere

stw

ith

adu

mm

yth

atis

equa

lto

one

whe

nth

e

indi

vidu

alis

the

head

ofth

eho

useh

old.

23

Table5:

Effects

ofdistan

cebe

tweencash

tran

sfer

andthereferenc

eweekon

working

forpa

idactiv

ities

All

Recipients

Recipients’

partner

Tran

sfer

was:

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

One

weekbe

fore

thereferenceweek

0.027

0.070

0.021

-0.019

-0.015

-0.013

0.004

0.028

0.028

(0.035)

(0.065)

(0.036)

(0.037)

(0.062)

(0.041)

(0.016)

(0.024)

(0.024)


-0.020

-0.062

-0.046

-0.030

-0.086

-0.054

-0.026

-0.003

-0.003

(0.037)

(0.065)

(0.035)

(0.037)

(0.064)

(0.041)

(0.021)

(0.026)

(0.026)

Atleaston


0.005

-0.029

-0.043

0.001

-0.081

-0.047

-0.003

0.001

0.001

(0.034)

(0.065)

(0.035)

(0.035)

(0.065)

(0.042)

(0.016)

(0.019)

(0.019)

Mun

icipality

fixed

effects

No

Yes

Yes

No

Yes

Yes

No

Yes

Yes

Add

itiona

lcon

trols

No

No

Yes

No

No

Yes

No

No

Yes

Observation

s1,577

1,577

1,577

827

827

827

750

750

750

R-squ

ared

0.001

0.047

0.720

0.001

0.358

0.699

0.004

0.402

0.402

Not

e:C

lust

ered

stan

dard

erro

rsat

the

vill

age

leve

lin

pare

nthe

ses.

Add

itio

nal

cont

rols

incl

ude:

sex,

mar

ital

stat

us,

age,

educ

atio

n,na

tive

lang

uage

indi

cato

rs,

pov

erty

stat

usdu

mm

ies

and

adu

mm

yva

riab

leth

atin

dica

tes

whe

ther

orno

tth

ein

divi

dual

live

sin

aru

ral

area

.F

inal

lyw

ein

tera

ctou

rdu

mm

ies

ofin

tere

stw

ith

adu

mm

yth

atis

equa

lto

one

whe

nth

e

indi

vidu

alis

the

head

ofth

eho

useh

old.

24

Table 6: Effects of distance between cash transfer and the reference week on daily hours of work(recipients only)

Sunday Monday Tuesday Wednesday Thursday Friday Saturday

Transfer was: (1) (2) (3) (4) (5) (6) (7)One week before the reference week 0.828* -0.732 -0.810* -1.269** -1.319*** -1.334*** -0.982*

(0.452) (0.494) (0.491) (0.498) (0.479) (0.514) (0.535)During the reference week 1.231** -0.183 -0.760 -0.783 -0.994* -0.644 0.297

(0.500) (0.531) (0.527) (0.520) (0.534) (0.548) (0.552)At least one week after the reference week 1.341*** 0.342 0.225 0.019 -0.084 0.074 0.656

(0.487) (0.453) (0.458) (0.470) (0.493) (0.459) (0.533)

Municipality fixed effects Yes Yes Yes Yes Yes Yes YesAdditional controls Yes Yes Yes Yes Yes Yes Yes

Observations 827 827 827 827 827 827 827R-squared 0.398 0.366 0.395 0.373 0.366 0.379 0.345

Note: Clustered standard errors at the village level in parentheses. Additional controls include: sex, marital status, age, education, native

language indicators, poverty status dummies and a dummy variable that indicates whether or not the individual lives in a rural area. Finally

we interact our dummies of interest with a dummy that is equal to one when the individual is the head of the household.

25

Table 7: Effects of distance between cash transfer and the reference week on weekly hours ofwork by Marital Status

Married Not Married

Transfer was: (1) (2) (3) (4)One week before the reference week -2.318 -11.293*** -2.825 -2.620

(1.875) (3.667) (2.068) (4.647)During the reference week -2.165 -6.788 0.088 3.899

(2.058) (4.599) (2.325) (4.557)At least one week after the reference week -1.310 -1.808 3.246 5.252

(1.745) (3.823) (2.023) (4.344)

Municipality fixed effects No Yes No YesAdditional Controls Yes Yes Yes Yes

Observations 446 446 381 381R-squared 0.072 0.540 0.063 0.540

Note: Clustered standard errors at the village level in parentheses. Additional controls include: sex, age, education, native language

indicators, poverty status dummies and a dummy variable that indicates whether or not the individual lives in a rural area. Finally we

interact our dummies of interest with a dummy that is equal to one when the individual is the head of the household.

Table 8: Effects of distance between cash transfer and the reference week on weekly hours ofwork by Group Age

Young (Under 40) Old

Transfer was: (1) (2) (3) (4)One week before the reference week -4.224** -11.985*** -0.502 -2.088

(2.035) (4.350) (1.919) (4.887)During the reference week -3.163 -8.396* 0.841 4.143

(2.211) (4.671) (2.186) (4.883)At least one week after the reference week -0.897 -2.033 1.881 7.796

(1.845) (3.750) (1.893) (4.814)






26

Table 9: Effects of distance between cash transfer and the reference week on weekly hours ofwork by Children’s age

With Children aged 5 or less With children aged 6 or more

Transfer was: (1) (2) (3) (4)One week before the reference week -4.346** -9.962** -0.637 -6.152

(1.822) (3.878) (2.209) (4.867)During the reference week -0.336 -3.830 -2.747 -2.149

(1.903) (4.900) (2.664) (5.689)At least one week after the reference week 0.686 1.870 -0.624 4.646

(1.618) (3.548) (2.256) (5.791)






Table 10: Effects of distance between cash transfer and the reference week on weekly hours ofwork (recipients only, including non-poors)

Transfer was: (1) (2) (3)

One week before the reference week -2.632** -4.881** -5.998**(1.329) (2.466) (2.408)

During the reference week -1.564 -0.685 -1.948(1.319) (2.583) (2.597)

At least one week after the reference week 1.024 3.071 2.251(1.225) (2.251) (2.275)

Municipality fixed effects No Yes YesAdditional controls No No Yes

Observations 1,015 1,015 1,015R-squared 0.009 0.340 0.354




27

Table 11: Effects of distance between cash transfer and the reference week on weekly hours ofwork (non-beneficiaries housewifes)

Transfer was: (1) (2) (3)

One week before the reference week -0.054 -0.799 -0.210(1.542) (3.021) ( 3.024)

During the reference week -1.603 -2.155 -1.997( 1.557) (2.831) (2.793)

At least one week after the reference week 3.352 2.211 2.099( 1.663) (2.848) ( 2.790)

Municipality fixed effects No Yes YesAdditional controls No No Yes

Observations 927 927 927R-squared 0.010 0.333 0.348




28

Table 12: Effects of distance between cash transfer and the reference week on weekly hours ofwork using the last payment date (recipients only)

Poors Poors and Non-poors

Transfer was: (1) (2) (3) (4) (5) (6)One week before the reference week -1.229 -3.781 -4.847** -2.118 -5.194* -5.970**

(1.646) (3.171) (2.461) (1.570) (2.983) (2.905)During the reference week -1.049 1.595 0.817 -1.503 0.803 -0.094

(1.612) (3.344) (3.437) (1.506) (2.957) (3.016)At least one week after the reference week -0.951 2.238 1.550 -0.450 1.538 1.044

(1.346) (2.718) (2.718) (1.287) (2.536) (2.520)

Municipality fixed effects No Yes Yes No Yes YesAdditional controls No No Yes No No Yes

Observations 827 827 827 1,015 1,015 1,015R-squared 0.001 0.392 0.392 0.003 0.339 0.339




Table 13: Effects of distance between cash transfer and the reference week on weekly hours ofwork by Payment Mechanism

Payment Mechanism Bank Armored Van

Transfer was: (1) (2) (3) (4)One week before the reference week -3.195* 0.913 -2.335 -8.807***

(1.839) (4.353) (2.087) (3.056)During the reference week -0.670 7.421 -2.835 -6.207

(2.275) (4.529) (2.119) (3.864)At least one week after the reference week 0.201 11.202* -0.418 -1.006

(2.082) (6.448) (1.763) (2.667)






29

Documents

Centro de Estudios Distributivos, Laborales y Sociales · Distributivos, Laborales y Sociales Maestría en Economía Facultad de Ciencias Económicas Conditional Cash Transfers, Payment